Hello! I was wondering if it is possible to directly access the flat address of a thread, as this shows when debugging in nsight (if I remember correctly). It would be particularly useful in multi-dimensional blocks, for example, I need to do a local reduction on some data in shared memory, where this reduction is only part of some larger computation for which is natural to have multi-dimensional blocks, however it seems wasteful to me to have to recompute the flat address (natural for the reduction) every time, so is there a way to access this just like we can access threadIdx.x?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Are 3D grids possible? Beginner question | 3 | 2227 | May 15, 2007 | |
| block and thread IDs? | 2 | 2238 | April 20, 2008 | |
| physical locations of the thread blocks or threads | 3 | 1693 | November 1, 2008 | |
| Access to CUDA Shared memory from the host | 4 | 1186 | December 18, 2018 | |
| bloking while reading from texture memory | 1 | 612 | December 28, 2014 | |
| Whether this is coalescing access several cases to decide | 0 | 1562 | August 2, 2011 | |
| Using CUDA to obtain physical address of texture? | 1 | 4570 | July 1, 2008 | |
| shared memory example | 1 | 5265 | February 24, 2010 | |
| Accessing a memory block created in another program. | 2 | 896 | September 7, 2018 | |
| Shared memory : shared access | 4 | 2021 | July 21, 2008 |